A Partitioning Tool for Designing Low-Power Partitioned Finite-State Machines

Various methods have been proposed to reduce Finite State Machines' power consumption (FSMs). Among them, an exciting solution is "partitioning", or "decomposition", which consists of the division of the original FSM into "N" sub-FSMs. While only one Sub-FSM receives the clock signal, the others are disabled, minimizing power consumption. However, considering large FSMs (sometimes more than 1,000 states), this method shows a high complexity of implementation, demanding some specific computational tool to perform automatic synthesis with reliability and accuracy. In this context, this paper presents a dedicated tool called "LSPart" that achieves and optimizes not only the "partitioning of FSMs" but also the "state assignment of each Sub-FSMs", using genetic algorithms. Asynchronous control is designed to switch the clock signal from one sub-FSM to the others, and an output block synchronizes sub-FSM outputs in a single one. Results show that this dedicated computational tool achieves excellent and reliable results in all studied cases, allowing the design of Low-Power Synchronous FSMs with power reduction of up to 82% compared to the original ones. Additional advantages of the method lead to the simplicity of handling the tool, even for extensive benchmarks, a single FSM output signal, and an optimal latency time, suggesting a high potential for practical implementation in actual low-power synchronous designs.